For a long time, we see here and there that there exist some racist translation machines and questions about “are machine translators racists?” Above all we see Google Translate in the list of racists translation machines. So I decided to look into the matter and gather some source to prove racists translation machines right or wrong! We try to answer one simple question: are machine translators racists? And if yes, where do they get their racism?!
Racist translation submissions on Google product forums
Let’s get started with racist translation submissions on Google product forums itself. Gabriel Bistriceanuser on Google product forums uses this phrase:
“Romanians are NOT GYPSIES” […] FIX IT IMMEDIATELY!
This user provides an attachment to confirms this claim that Google actually translates it otherwise!:
Valcea Ostroveni is another user that mentions the very same racist translation of Google about Gypsies and Romanians.
Another user, Taxlady, this time in Swedish to English translation kindly suggest that:
I have noticed that in translations of Swedish to English, if the word is anything to do with the Saami people, that the translation is often racist or insulting. Is there a way to report this? How does that happen? Is it a bunch of racist contributors putting racist translations for words? It happens in Norwegian too.
Amar Fox, another user, tests and thinks what we think! Personally, I think this is the most important issue that Amar Fox pointed out perfectly and we saw several other mentions about this kind of translation that is related to political agendas. He started his thread with this headline: ‘Why Google translate development team did this?’ and indeed that’s our question too regarding the following examples drawn out by Amar Fox:
Americans are terrorists – translate in arabic means “Americans are not terrorists”
Russians are terrorists – translate in arabic means “Russians are not terrorists”
Arabs are terrorists – translate in arabic means “Arabs are the terrorists or terrorism”
What about answers? Well, all of these are actually threads and questions that users have from Google Translate! Answers to all of these questions revolve around ‘submitting an edit’ (and being hopeful that Google will fix it) or some advanced users escalate threads to get more attention! But this doesn’t answer our question: Why there exist such translations in the first place? What if no one notices them? These aren’t just simple ‘errors’ to be fixed by user suggested edits, these are called and labeled ‘racist translation’.
An investigation of racist translation machines on other media
The Agenda reports (Politico) that St. Louis tech executive named Emre Şarbak was translating Turkish phrases into English. In Turkish Language there is single neutral pronoun instead of ‘he’ or ‘she’ in English, but when translating these phrases in Google Translate, ungendered Turkish sentence “o is a nurse” would become “she is a nurse,” while “o is a doctor” would become “he is a doctor.”
Quartz is another website that reports about gender bias that exists in Google Translate when translating from languages which use neutral pronouns as in Turkish and in some cases Chinese. Quartz reports that “when Google Translate goes from Turkish to English, it just has to guess whether “o” means he, she, or it. And those translations reveal the algorithm’s gender bias.” This website doesn’t stop here and composes a beautiful poem that is “on the topic of gender.” We suggest reading it for yourself! Four lines of this poem go like this:
he is a soldier
she’s a teacher
he is a doctor
she is a nurse
he is hardworking
she is lazy
Beautiful, right? And you can guess the rest!
Racist translation of WeChat, all over the news!
It’s not only in Google Translate that we see a biased and racist translation, China’s popular app, WeChat hit the headline of many news websites for translating ‘hei laowai’ into something far more offensive ‘the n-word’ rather than neutral ‘black foreigner’. Of course, they had the courage to apologize and sent Sixth Tone the following apology, without any further explanation:
“We’re very sorry for the inappropriate translation. After receiving users’ feedback, we immediately fixed the problem.”
In this snapshot from Sixth Tone we can clearly see what they are talking about:
Apparently, this came as a surprise to the authorities of this app too that why such a word even exists in the translator and why it is behaving like a racist translation machine. The spokesperson from WeChat explains that WeChat uses neural machine translation and its engine constantly gets refined to provide “more accurate, faithful, expressive, and elegant” results.
To this end we can’t make a clear answer to our primary and simple question: Are machine translators racists? To get an answer, we should get into the underlying level of machine translators.
Racist translation machines’ use of artificial intelligence
To get really deep into this matter, we should seek an answer on another level. We should seek the answer on how machine translation works and does it make machine translators racists? Does something make machine translators be racist translation machines? and more specifically, are machine translators racist per se?
Everybody knows that nowadays virtually all machine translators use artificial intelligence to do business. Now, we think that ‘machine’ can be just and even in face of any human being, but in reality, we see it otherwise. Maybe machines themselves can be free of prejudice, maybe it is the way machines get trained that makes them racists machines.
Google’s Translate, like many other automated machine translation tools, learns language from an existing corpus that often includes cultural patterns about the description of men and women. So, as you know, this written model often has biases on its own! as a result, any translation machine that learns from this biased model will actually be a racists translation machine! So after all, we cannot say that a piece of machine can be unbiased because its source of learning is actually biased. I saw a great headline for this matter on The Guardian that reads Robots are racist and sexist. Just like the people who created them that perfectly outlines this phenomenon that is mostly called “algorithmic bias”.
So in a way, we can say that it is not the machine’s fault that it is a racists translation machine. Their algorithm is based on a huge corpus of human language, so we should blame humans for feeding this kind of biased corpora to these machines.
Ofer Tirosh points out:
The problem seems to be that machine learning of this nature is undertaken using a vast array of texts written by humans. Those texts include countless instances of racism, sexism, ageism and pretty much any other -ism imaginable, both in their attitudes and in the language they use. As such, we’re training machine learning systems to incorporate just the same prejudices and biases as the human brain is capable of producing. At a basic level, machines learn from the data that we put into them. If that data is flawed, then the resulting program will reflect those flaws.
Are Machine Translators Racists?
Let’s get back to our simple and primary question: Are machine translators racists? A simple answer would be they are not racists translator machines! it is their source that makes them racists and biased. We still need human translation services that are more unbiased than those of machines. We previously talked about how a judged used google translate on a pre-trial hearing that now comes as more surprising to us.
But what is your idea? Are machine translators racists? Do you think a human translation is more unbiased than a machine translation? Or not? Let us know! Consider contributing to Dilmanj and publish your thoughts on Dilmanj.